rcurtin on FreenodeI guess because there are too many columns and they wrap?
shritActually there is about 10K rows and three columns
rcurtin on Freenodeoh, maybe print with .t()?
rcurtin on Freenodebut I'm not totally sure what you mean, maybe paste the output?
shrit@rcurtin, another question considering the meaning of the word ´coordinate´ in ensmallen. If I use ensmallen optimizer to train a neural network the word coordinate means the weights of the netowrk, am I right?
shritExactly, So If I want to store the best coordinate, I will have to use the callback ens::StoreBestCoordinate<> and then these coordinates are stored to a specific matrix, I want just to ask how to be sure that the Best coordinates are saved into the model.bin at the end?
shritI will do a pull request and update an existing example, this one worth to be noted down, many thanks :+1
travis-ci on Freenodeshrit/examples#99 (best_coordinates - 2bb362b : Omar Shrit): The build passed.
travis-ci on Freenodeshrit/examples#101 (fix_mnist_simple_bug - 238aff2 : shrit): The build passed.
rcurtin on FreenodeI didn't add any of the catch or WeightSize refactorings, since those aren't actually user visible (so if they get in, great, if not, it shouldn't matter for users)
rcurtin on Freenodeif anyone else has any PRs that are "just about" done and can be merged in the next couple of days, feel free to point them out or just add them to the milestone! :)
abernauerYeah the latter is easy enouth
shritThe difference is only the inclusion of some layers + FFN headers, there not even ensmallen code nor optimizer
shritI think that boost visitors are taking a considerable amount of time, there is no way that template usage in FFN code to consume such amount of time.
There is a test in fastmks_test.cpp at line 180 which fails with this expansion:
REQUIRE( pk.Evaluate(denseset.col(i), denseset.col(j)) == 0.0 ) with expansion: 0.0 == 0.0
I tried to use
Approx(0.0) then it was failing with
0.0 == Approx(0.0) and even tried
Approx(0.0).epsilon(1e-10) it was still failing with
0.0 == Approx(0.0)
I don't seem to get how getting
0.0 == 0.0 fails the test ?
travis-ci on Freenodemlpack/ensmallen#1066 (master - 25d0d84 : Marcus Edel): The build was fixed.